Google: StarCraft is likely to be the next target for DeepMind

After beating the best human go player by DeepMinds AlphaGo, the next challenge from Google’s business is likely to be StarCraft, DeepMinds founder Demis Hassabis said in an interview.

In his conversation with The Verge, he does indicate that the AlphaGo program is a project within a much larger story in which Google ‘wants to solve the issue of intelligence’. AlphaGo has shown that it has been able to beat the best human player in a so-called perfect information game or game where all information is available. On go, all information is available to both players by looking at the game board. Go is generally considered the hardest for artificial intelligence in that category.

A game like poker is a game in which information is partly available, but in which a piece of information is unknown, also known as an imperfect information game. Also, poker involves more than two players in total, which also increases the difficulty.

Within the world of computer games, StarCraft is probably the next challenge for the team behind AlphaGo. At least, that’s what Hassabis thinks. The comment is corroborated by Google Senior Fellow Jeff Dean in Business Insider.

Still, beating players in StarCraft is not an end in itself. “We’re only interested in things that fall within the research program. So the goal isn’t to have DeepMind beat people in games, no matter how fun and exciting that is.”

In other words, Hassabis wants to use games as long as they are useful as a testing platform to test ideas around algorithms and how far they scale. Using games for that is one of the ways to do that efficiently. Ultimately, the systems must be deployed for real-world problems.